Is the Diversification Discount Caused by the Book Value Bias of Debt? Abstract We analyze whether the diversification discount is driven by the book value bias of debt. In contrast to the related study by Mansi and Reeb (2002), who use actual market values of debt instead of book values, we replace the book value of debt by market value of debt estimates that are based on often used implementations of the Merton (1974) model. This procedure is possible for almost all firms in our sample of all German non-financial CDAX firms from 2000 to 2006. This is the main advantage when compared to the Mansi and Reeb (2002) study who have access to market values of debt for only about 13% of U.S. firms in their sample. We are thus able to circumvent one weakness of the Mansi and Reeb (2002) study. Using the standard excess value approach suggested by Berger and Ofek (1995), we find that diversified firms trade at a significant discount of about 15% from 2000 to 2006, which is the first period in which German firms have to disclose segment information comparable to U.S. accounting rules. However, we find evidence that the use of book values of corporate debt in the computation of the excess value underestimates the firm value of diversified firms compared to focused ones. In line with Mansi and Reeb (2002), the diversification discount is reduced but does in contrast to their evidence - not completely vanish. This result hints to the point that the results presented in Mansi and Reeb (2002) might be driven by sample selection bias or lack of statistical power. We conclude that the book value of debt bias is only one explanation for the diversification discount. Keywords: diversification, diversification discount, conglomerate discount, internal capital markets, option pricing, debt valuation, Merton model JEL Classification Code: G12, G13, G14, G31 2 1 Introduction The benefits and costs of corporate diversification have been the subject of extensive empirical and theoretical research.1 Weston (1970) suggests that diversified firms have the ability to use economies of scale because they provide more efficient operations and more profitable lines of business when compared to stand-alone firms. Lewellen (1971) argues that diversified firms enjoy greater debt capacity and debt tax shields relative to singleline firms due to lower firm risk. Furthermore, diversification can create internal capital markets that enable firms to pool and reallocate corporate resources more efficiently through “winner picking” than through external financing, which may increase investment efficiency (see e.g. Stein (1997)). The negative impacts of corporate diversification are often described in terms of inefficient investments due to cross-subsidization between divisions. Rajan, Servaes, and Zingales (2000), for example, model the distortions that internal power struggles generate in resource allocations among the divisions of a diversified firm. Other costs of diversification include investments in lines of businesses with poor investment opportunities (e.g., Stulz (1990)). Jensen (1986) argues that diversified firms invest more in negative cash flow projects than their segments would if operated independently. Meyer, Milgrom, and Roberts (1992) reinforce this argument, suggesting that stand-alone segments tend to produce lower losses than diversified firms due to the cross-subsidization of business segments with negative cash flows. Scharfstein and Stein (2000) suggest that rent-seeking behavior by divisional managers undermines the functioning of internal capital markets and leads to inefficient investments. Overall, the published literature on corporate diversification suggests that conglomeration is associated with greater agency costs (see also Jensen (1993)). These agency costs are manifested in the form of accepting negative net present value projects, which reduce the value of the firm. The empirical literature mainly documents a so-called “diversification discount”. As developed by Lang and Stulz (1994) and Berger and Ofek (1995), the diversification discount compares the market value of a diversified firm to the imputed stand-alone values of its 1 Recent surveys of this literature are Hellwig, Laux, and Müller (2002), Maksimovic and Phillips (2007), Stein (2003), and Martin and Sayrak (2003). 3 individual segments. These imputed values are based on multiples (such as price-to-book value, or price-to-sales) of comparable pure-play firms in the same industries as the diversified firm’s segments. Using data from the U.S., these authors find substantial mean discounts, on the order of 15%, which they interpret as evidence of value destruction by diversified firms. This work has been extended to a variety of other sample periods and countries by Servaes (1996), Lins and Servaes (1999), or Lins and Servaes (2002). Results suggest that the diversification discount is a pervasive phenomenon. However, a number of other papers cast doubt on the interpretation that the diversification discount reflects value destruction. Campa and Kedia (2002), Chevalier (2004), Graham, Lemmon, and Wolf (2002), Hyland and Diltz (2002), and Villalonga (2004) all argue in one way or another that the discount is tainted by endogeneity bias, because relatively weak firms are the ones that choose to diversify in the first place. A balanced reading of these papers suggests that taking these caveats into account significantly reduces - though may not eliminate - the discount. More recent papers try to analyze the determinants of the large cross-sectional variation of the diversification discount and its variability over time and across countries. Santalo and Becerra (2008), for example, show that the effect of diversification on performance is not homogeneous across industries and explore analytically and empirically the implications of this finding for the diversification literature. Diversified firms perform better in industries with a small number of non-diversified competitors or, equivalently, when specialized firms have a small combined market share, but worse as the presence of specialized firms increases in the industries in which they compete. Lyandres (2007) proposes another explanation for the large cross-sectional variation in the excess values of diversified firms. His model applies the idea of shareholders’ limited liability affecting firms’ output market strategies to the analysis of financial and operating choices of conglomerates. The inability of conglomerates to flexibly choosing their divisions’ capital structures, reduces their value. Thus, the model highlights one type of inefficiency of the conglomerate organizational structure, which is suboptimal financing. Lyandres (2007) supports his predictions in an empirical part. Borghesi, Houston, and Naranjo (2007) examine corporate product diversification as a dynamic process. They find that a significant portion of the diversification discount arises from benchmark comparisons of value ratios of mature 4 firms with those of very young firms that are more likely to have high value multiples. The magnitude of the diversification discount is lower by 15% to 30% (i.e., the excess value is higher) when they control for firm age. Borghesi, Houston, and Naranjo (2007) also show that diversification reduces the mortality rate of firms, and they provide evidence that mature firms pursue diversification strategies partly as a means to exit stagnant business segments for industries that are more highly valued. Lee, Peng, and Lee (2008) argue that a diversification premium in emerging economies is likely to dissipate over time and eventually become a diversification discount. They empirically draw on a data set from South Korea and document the longitudinal process of how a diversification premium becomes a diversification discount during institutional transitions. Overall, the diversification discount seems to be such a stable fact that consulting firms base their strategy suggestions on these findings. For example, Boston Consulting Group (2006) writes how diversified firms can create value. Even textbooks pick up the arguments of the early literature and state that the diversification discount is likely to be the consequence of agency problems and inefficient resource allocation in conglomerates.2 Given the vast amount of literature it is surprising that one weakness of the Berger and Ofek (1995) excess value measure is hardly addressed: the risk effects of conglomeration and its subsequent impact on firm value. While diversification reduces shareholder value, it enhances bondholder value due to a reduction in firm risk. Mansi and Reeb (2002) is the only published study which makes this point and empirically investigates the impact of conglomeration on firm value. They obtain the market values of both debt and equity for a subset of U.S. firms and examine the bias in using book values of debt to compute excess values. Consistent with the hypothesis that diversification leads to lower firm risk, they find that book values of debt are a more downward biased proxy of the market value of debt for diversified firms, relative to undiversified firms. This finding suggests that measures of firm value based on book values of debt systematically undervalue diversified firms. When considering the joint impact of diversification on debt and equity holders, they find that, on average, corporate diversification is insignificantly related to excess firm value. Their conclusion is that diversification reduces shareholder value, increases bondholder value, and has no impact on total firm value. 2 See e.g. Hill and Jones (2007). 5 The fact that Mansi and Reeb (2002) is the only published empirical study dealing with the risk effect of corporate diversification and its impact on firm value that we are aware of, is even more surprising given that several theoretical papers examine the consequences of corporate diversification explicitly assuming that it leads to lower firm risk. As such, Lewellen (1971) argues that diversified companies enjoy higher debt capacities because their default risk is lower. As a consequence the value of the company’s tax shield increases which enhances the company’s overall value as well. Amihud and Lev (1981) argue that managers engage in corporate diversification, even if it reduces shareholder value, to reduce their human capital risk. The presumption is that corporate diversification lowers firm risk. In a contingent claims framework, lowering firm risk should lower shareholder value and increase bondholder value. While our study is closely related to the research of Mansi and Reeb (2002), we propose a different way to incorporate market values of debt which circumvents a weakness of the Mansi and Reeb (2002) study: They collect market values only for a small subset (13%) of their original sample of all U.S. firms, which could lead to a potential selection bias. For instance, one could assume that larger and, on average, more successful companies have more traded debt outstanding (thereby excluding all the poor diversifiers) because they are more likely to be given a good rating by a rating agency. In addition, Mansi and Reeb (2002) have to compute imputed firm values using book values of debt due to the small size of their sub-sample. Furthermore, the coefficient on the multi-segment dummy in their regression specification is of similar size (-0.117) as the one in Berger and Ofek (1995), but the coefficient is insignificant. Thus, their conclusion that the observed diversification discount in previous studies represents the wealth transfer from shareholders to bondholders and not the reduction in firm value might result from a lack of statistical power. In contrast, we employ several specifications of the Merton (1974) bond pricing model which were previously used in different research contexts to estimate the market value of the firm’s assets. Our estimation procedure, which will be described in detail below, requires only very little additional information and thus can be done for almost all focused and diversified companies for which an excess value based on debt book values can be calculated. Moreover, even if one has access to a research database to infer the market 6 price of debt, most corporate debt is not traded as mentioned above. This is especially true for bank-based systems like Germany. In this case one either has to use estimates of market values or to rely on book values. Eberhart (2005) shows that applications of the Merton (1974) model provide more accurate debt value estimates than the book value approximation. Moreover, the estimation procedure offers several important advantages compared to the collection of market prices of corporate debt. Most importantly, it is difficult (and perhaps tedious) to obtain this data.3 Our study is related to one recent working paper. Ammann, Hoechle, and Schmid (2008) treat the entire long-term debt on the books of firms as one coupon bond with the coupon set equal to the interest expenses on all debt. They then value this coupon bond at the current cost of debt for the company approximated by the yield of a bond portfolio with the same credit rating. As Compustat provides an official credit rating from S&P only for a very limited subset of their sample, they alternatively construct an artificial credit rating based on the interest coverage ratio. Their sample consists of all firms with data reported on both the Compustat Industrial Annual and Segment data files and covers the period from 1998 to 2005. Ammann, Hoechle, and Schmid (2008) show with their crude approximation of the market value of debt that the potential effect of accounting for differences between the market and book value of debt on the conglomerate discount in the U.S. is very limited. Our main findings can be summarized as follows. In a first step, we document that German conglomerates trade at a significant discount of 15% when the traditional Berger and Ofek (1995) measure is used. We find that our different excess value measures are highly significantly positively correlated. In line with Mansi and Reeb (2002), the diversification discount is reduced but does - in contrast to their evidence - not completely vanish. This hints to the point that the results presented in Mansi and Reeb (2002) might be driven by sample selection bias and lack of statistical power. We conclude that the book value of debt bias is only one explanation for the diversification discount. The remainder of the study is organized as follows. Section 2 contains a sketch of the Merton (1974) model, a discussion of its usage in research and practice as well as the general procedure to estimate market values of debt. In Section 3, we describe the data set, the identification of focused and diversified firms, and the excess value measure. In 3 Mansi and Reeb (2002) use the Lehman Brothers Fixed Income database in their study. 7 Section 4, we outline the precise procedure of how we estimate the market values of debt. Furthermore, we assess the quality of our estimation by comparing market value of debt estimates with actual bond prices for a subset of our initial firm sample. Section 5 presents the results and the last section concludes. 2 Estimating Market Values of Debt Using the Merton Model 2.1 Sketch of the Model In this subsection we describe the Merton (1974) bond pricing model.4 The Merton (1974) model makes the following assumptions. The first is that the total value of a firm follows a geometric Brownian motion, dV = µV dt + σV V dW, (1) where V is the total value of the firm, µ is the expected continuously compounded return on V , σV is the volatility of firm value and dW is a standard Wiener process. The second assumption of the Merton (1974) model is that the firm has issued just one discount bond maturing in T periods. Under these assumptions, the equity of the firm is a call option on the underlying value of the firm with a strike price equal to the face value of the firm’s debt and a time-to-maturity of T . Moreover, the value of equity as a function of the total value of the firm can be described by the Black-Scholes-Merton formula. By put-call parity, the value of the firm’s debt is equal to the value of a risk-free discount bond minus the value of a put option written on the firm with a strike price equal to the face value of debt and a time-to-maturity of T . Symbolically, the Merton model states that the equity value of a firm satisfies E = V N (d1 ) − e−rT F N (d2 ), (2) where E is the market value of the firm’s equity, F is the face value of the firm’s debt, 4 The following paragraphs are based on Bharath and Shumway (2008). 8 r is the instantaneous risk-free rate, N (·) is the cumulative standard normal distribution function, d1 is given by d1 = and d2 is d1 − σV √ ln ³ ´ V F + (r + 0.5σV2 )T √ , σV T (3) T. The Merton (1974) model makes use of two equations. The first is the Black-ScholesMerton Equation (2), expressing the value of a firm’s equity as a function of the value of the firm. The second relates the volatility of the firm’s value to the volatility of its equity. Under the assumptions in Merton (1974), the value of equity is a function of the value of the firm and time, so it follows directly from Ito’s lemma that µ σE = V E ¶ ∂E σV . ∂V In the Black-Scholes-Merton model, it can be shown that (4) ∂E ∂V = N (d1 ), so that under the assumptions in the Merton (1974) model, the asset volatility of the firm and its equity volatility are related by µ ¶ V σE = N (d1 )σV , E (5) where d1 is defined as in Equation (3). In the Merton (1974) model, the value of the option is observed as the total value of the firm’s equity, while the value of the underlying asset (the total value of the firm) is not directly observable. Thus, while V must be inferred, E is easy to observe in the marketplace by multiplying the firm’s shares outstanding by its current stock price. Similarly, in the Merton (1974) model, the volatility of equity, σE , can be estimated but the volatility of the underlying firm, σV , must be inferred. To obtain the market value estimates of debt, one simply has to subtract the market capitalization E from the the inferred total firm value V . The first step in implementing the Merton (1974) model is to estimate σE from either 9 historical stock returns data or from option-implied volatility data. The second step is to choose a forecasting horizon and a measure of the face value of the firm’s debt. For example, it is common to use historical returns data to estimate σE , assume a forecasting horizon of 1 year (T = 1), and take the book value of the firm’s total liabilities to be the face value of the firm’s debt.5 The third step is to collect values of the risk-free rate and the market equity of the firm. After performing these three steps, we have values for each of the variables in Equations (2) and (5) except for V and σV , the total value of the firm and the volatility of the firm value, respectively. The fourth step in implementing the model is to solve Equations (2) and (5) numerically for values of V and σV . In the next section, we describe the exact procedure of how we estimate the market values of debt and which parameter values we use. 2.2 Application of the Merton Model in Research and Practice Bharath and Shumway (2008) state that the Merton (1974) model is a rather unusual forecasting model. Most forecasting models constructed by econometricians involve posing a model and then estimating the model with method of moments or maximum-likelihood techniques. The Merton (1974) model actually involves no estimation. Instead, it replaces estimation with something more like calibration-solving for implied parameter values. There are wide spread applications of the Merton (1974) model such as pricing of credit risk (see e.g. Duffie and Singleton (2003)) or forecasting of corporate default (see e.g. Bharath and Shumway (2008)).6 A popular implementation of the model is the commercial KMV model (see Crosbie and Bohn (2003)).7 5 This simplifying assumption is, for example, used by Bharath and Shumway (2008), Campbell, Hilscher, and Szilagyi (2008), or Crosbie and Bohn (2003). However, we will also analyze different parametrizations of the model. We, e.g., deviate from the T = 1 assumption in our analysis presented in the following. 6 Other recent studies which use the Merton (1974) model are Duffie, Saita, and Wang (2007) and Campbell, Hilscher, and Szilagyi (2008). 7 Moody’s KMV is the world’s leading provider of quantitative credit analysis tools to lenders, investors, and corporations (www.moodyskmv.com). 10 3 Construction of the Data Set 3.1 Sample of Firms and Data Set The starting point of our analysis is the sample of all CDAX firms from 2000 to 2006. The reason for starting in the year 2000 is that from this year on reliable business segment data is available in Germany. To construct our final data set, we proceed as follows. From Datastream/Worldscope, we download business segment information for all German (financial and non-financial) firms that have been members of the CDAX index in at least one year during the period 2000 and 2006. The CDAX covers all German firms whose shares were admitted to the Prime Standard and General Standard segments of the German Stock Exchange. Before the introduction of these segments in the year 2003, the CDAX contained all firms whose shares were members of the segments Amtlicher Handel, Geregelter Markt, and Neuer Markt.8 The CDAX index reflects the full spectrum of the German equity market, and is consequently well suited for academic purposes. To identify the firms in the CDAX, we obtain end of year lists on the index members from the Karlsruher Kapitalmarktdatenbank (KKMDB) maintained by the University of Karlsruhe in Germany.9 The number of firms listed in the CDAX during the sample period varies between 678 and 790 (see Table 1). The number of firms generally decreases over time indicating that several firms were delisted after the stock market boom around the year 2001. In 2006, the number of listed firms slightly increases. For our empirical analysis, we clean the data set in the following way. Some firms appear twice in our CDAX list as they are listed with two types of stocks (common stock and preferred stock).10 Therefore, we delete all duplicate observations. Firms in the remaining data set are identified by their ISIN of the common stock. A firm is categorized as a firm with business segment information if one of the following variables is available for at least one (operating or non-operating) segment: segment assets (Worldscope data item wc19503), segment sales (wc19501), or segment description 8 See http://deutsche-boerse.com for details. 9 see http://fmi.fbv.uni-karlsruhe.de/149.php for details. 10 Some firms have even more than two types of shares. 11 (wc19500). To construct our final data set, we merge the business segment data with the complete CDAX list. We then drop all financial firms as their balance sheet data is not comparable to non-financial firms. We identify financial firms by their primary SIC code (Worldscope data item wc07021) and delete all firms with SIC codes between 6000 and 6999. Table 1 presents the number of firms for which we have at least one piece of business segment information during the sample period. 3.2 Identification of Operating Segments and Classification Procedure to Obtain Focused and Diversified Firms We classify all firms without business segment information as focused firms. We classify a firm as diversified if the number of operating segments which operate in different 2-digit SIC codes is 2 or higher.11 All other firms are also classified as focused. For this classification, we disregard all non-operating segments. We define a segment as non-operating if one of the following criteria is fulfilled: - The segment description (Worldscope code wc19500 “Product Segment 1 - Description” to 19590 “Product Segment 10 - Description”) contains strings which indicate that the segment is a non-operating segment.12 - The segment SIC code provided by Worldscope (Worldscope code wc19506 “Product Segment 1 - SIC Code” to wc19596 “Product Segment 10 - SIC Code”) is 9999 (nonclassifiable establishments). - Segment assets (Worldscope code 19503 “Product Segment 1 - Assets” to wc19593 “Product Segment 10 - Assets”) are 0 or negative (because such segments can be regarded as adjustment segments). 11 This classification procedure is standard for non-U.S. countries. 12 Such strings are, for example, holding, central division, central services, corporate, corporate center, corporate services, other & holding, consolidated, consolidation, intercompany, inter-company, intergroup, intersegment, intra-group, intragroup, adjustment, unallocated, not allocated, transfer, administration. The search for these strings is case insensitive and spaces are ignored. I.e., we classify a segment as non-operating when the segment description contains the strings holding, Holding, central services, or centralservices, for example. 12 - Segment sales (Worldscope code wc19501 “Product Segment 1 - Sales” to wc19591 “Product Segment 10 - Sales”) are 0 or negative (because such segments can be regarded as adjustment segments). Whenever we refer to a focused firm, we mean (i) firms with only one operating segment, (ii) firms with more than one operating segment which all operate in the same two digit SIC code industry or (iii) firms without business segment information at all.13 Consistent with the literature (see, for example, Rajan, Servaes, and Zingales (2000), p. 54), we ensure that no data item is calculated using data spread over multiple years. One reason for this convention is that it is impossible to identify for a given business segment the respective business segment of the last year due to potential name changes or reorganizations. Furthermore, we collect data on several other variables which are described in Table 2. 3.3 Measuring the Excess Value The excess value of a company is the natural logarithm of the ratio of a firm’s actual value to its imputed value. A firm’s imputed value is the sum of the imputed values of its segments, with each segment’s imputed value being equal to the segment’s sales multiplied by its industry median ratio of total capital (market value of equity plus book value of debt or market value estimate of debt) to sales. More precisely, excess value EVi and imputed value I(V )i of company i, are defined as ! Ã Vi , and EVi = ln I(V )i I(V )i = n X AIij · (multiple of segment j of firm i), with (6) (7) j=1 13 All results are robust with respect to the definition of focused and diversified firms. Furthermore, the results are unchanged when we drop all firms without business segment information from our sample. 13 V = total capital (market value of equity plus book value of debt or market value estimate of debt) Multiple of segment j of firm i = median ratio of V to accounting item (sales ratio) of focused firms in industry of segment j AIij = accounting item of segment j of firm i n = number of segments of firm i Note that it is also possible to calculate the excess value measure for focused (single segment) firms, i.e. firms with “only one segment”. In other words, some focused firms may trade at a premium while others might trade at a discount. However, the median focused firms has, by construction, an excess value of 0. The contribution of this paper is that we not only calculate the traditional Berger and Ofek (1995) excess value measure in which the total capital of a firm is calculated as market value of equity plus book value of debt. We also substitute the book value of debt by market value of debt estimates that are based on implementations of the Merton (1974) model described in the previous subsection. 4 4.1 Estimation of Market Values of Debt Detailed Estimation Procedure for Market Values of Debt To estimate market values of debt, we solve equations (2) and (5), a system of nonlinear equations, for V and σV . Therefore, we need values for the remaining variables: Stock return (equity) volatility (σE ), time to maturity (T ), the risk-free rate (r), the face value of debt (F ), and the market capitalization (E). We use the set of input parameters suggested by three recent studies (Bharath and Shumway (2008), Eberhart (2005), and Vassalou and Xing (2004)). Table 3 shows a summary of the input parameters used in this paper to estimate market values of debt which are based on the above mentioned three papers. It is common to use historical returns data to estimate σE . Furthermore, several studies assume a forecasting horizon of 1 year (T = 1), and take the book value of the firm’s total 14 liabilities to be the face value of the firm’s debt. Assuming a time to maturity of T = 1 is quite common (see, e.g., Bharath and Shumway (2008), Campbell, Hilscher, and Szilagyi (2008), or Crosbie and Bohn (2003)). However, this assumption is an oversimplification. This is why we also use and prefer the Eberhart (2005) parametrization. He explicitly takes the amount of long-term debt into account. As an approximation of the time to expiration for the capital structure presumed in the Merton (1974) model, he estimates the weighted average of the duration for the firm’s short-term debt and long-term debt as 0.6·(short-term debt ratio) +6.3·(long-term debt ratio). Thus, Eberhart (2005) assumes a short-term debt duration of 0.6 years and long-term debt duration of 6.3 years.14 The next step is to collect values of the risk-free rate and the market equity of the firm. Then, we have values for each of the variables in the two equations except for V and σV , the total value of the firm and the volatility of firm value, respectively. The last step is to solve the system of nonlinear equations numerically to obtain the market value of debt estimate. 4.2 Quality of Estimation In the remainder of this subsection we address the question of how good the Merton (1974) model works. To do this, we first summarize evidence found in the literature. In a second step, we compare the market value estimates obtained by using the above parametrizations with real bond prices of (a subset of) our firm sample. Eberhart (2005) performs two series of tests comparing the Merton (1974) model to the book value of debt approximation. Using stock and bond data, he finds consistent evidence that the Merton (1974) model provides more accurate debt value estimates than the book value of debt approximation. Second, he shows that the Merton (1974) model is an easily estimated alternative to the book value of debt approximation as it only requires standard data available in data bases like CRSP or Compustat. Therefore, he concludes that the book value of debt approximation not only does create important biases, but it is also an unnecessary simplification. 14 Eberhart (2005) estimates the debt duration for short-term (long-term) debt using the weighted average durations of bonds with durations of 1 year or less (more than 1 year) in the Lehman corporate bond database. 15 Eom, Helwege, and Huang (2004) empirically test several structural models of corporate bond pricing, among them the Merton (1974) model. They implement the models using a sample of 182 bond prices from firms with simple capital structures during the period 1986 to 1997. They find that more sophisticated structural models of corporate bond pricing do not outperform the simplest model which is the Merton (1974) model. Schaefer and Strebulaev (2008) show that structural models of credit risk provide quite accurate predictions of the sensitivity of corporate bond returns to changes in the value of equity (hedge ratios). The main result of this paper is that even the simplest of the structural models, the Merton (1974) model, produces hedge ratios that are not rejected in time-series tests. To evaluate the accuracy of the debt market value estimates for the three different model parametrizations we obtain bond prices from the iBoxx EUR Corporates database for a subset of our firm sample. The data is provided by Deutsche Börse AG which calculates and disseminates the iBoxx EUR index family. Those indices aim at representing the investment grade fixed-income market for Euro-denominated bonds. They comprise all bonds issued in the Euro-zone by central and local governments, banks, and private corporations. A bond must have a predetermined cash flow structure (e.g. plain vanilla bonds or zero coupon bonds), a time to maturity of at least one year and a minimum amount outstanding of 500 million Euro to be included. Moreover, an investment grade rating for the bond must be provided by at least one rating agency (Moody’s, S&P, or Fitch). The iBoxx EUR Corporates database contains information such as market value, nominal amount, coupon, yield, maturity, and credit rating of the bonds. We obtain the constituent lists for the period from December 2000 to December 2006 and match them with our CDAX sample. While the iBoxx index series is biased towards larger bonds being rated, it is, to the best of our knowledge, the most comprehensive database on bond prices for Euro-denominated bonds. It is managed with a philosophy to reflect the investable fixed income universe and, as a result, market values of debt should not be distorted by illiquidity constraints. Table 4 shows the number of firms in our sample with tradable debt securities, separated 16 by year and diversification degree. Of the total CDAX universe only 8 firm-year observations classified as focused have bonds outstanding. In contrast, diversified firms are bond-financed to a larger extent as indicated by the 313 diversified firm-year observations with bond data. Still this number is rather small compared to the total firm universe. Moreover, inspection of Columns 2 and 3, which relate the amount outstanding to total debt and long-term debt, shows that most corporate debt in our sample is not traded. The average ratio is 13.6% for total debt and 20.0% for long-term debt. For bank-based systems like Germany, it is obviously necessary to estimate market values of debt if one does not want to rely on book values. In the following, we provide information on the accuracy of the three estimation methods (i.e. parametrizations). In order to do so, we compare the market-to-book ratios for total firm debt using the market value estimates with the observed ratios for the bonds outstanding. Table 5 depicts the cross-sectional variation in bond market values to face values and market-to-book ratios for total firm debt for all three estimation methods. There is a considerable variation in bond prices which highlights the potential benefit of using market value estimates instead of book values. While the mean bond market value to face value is 1.08, the highest value is 1.46 and the lowest value is 0.95.15 However, if we look at the relation of market value estimates to book values, we find that the Bharath and Shumway (2008) and Vassalou and Xing (2004) parametrizations are not capable of replicating this observed cross-sectional heterogeneity. For both models, minimum, mean, and maximum market-to-book ratios are very similar and close to 1. In contrast, the Eberhart (2005) model produces a dispersion in market-to-book ratios which is very similar when compared to the observed distribution of bond market-to-face value ratios. The mean ratio is 1.09, the minimum is 0.93 and the maximum is 1.38. In Table 6, we compare the estimation error based on the three different market estimates which is defined as follows: 15 The fact, that most bonds are traded above their face values may reflect the decline of the yield curve over the sample period as well as the investment-grade quality of the bond issuers. 17 Estimation Error = ( Actual M arket V alue Estimated M arket V alue )Bond − ( )F irm . F ace V alue Book V alue (8) Hence, we compare the market-to-book ratio based on observed bond prices with the market-to-book ratio based on market value estimates for each firm-year observation.16 We also calculate the estimation error replacing market value estimates with book values to assess the bias in using book values for firm debt and start with an examination of this issue in Column 1. Over the complete sample period there is an considerable book bias of 0.08 or 8%, indicating that book values of debt are a downward biased proxy for market values. The mean estimation error varies over time, but exists in all years. However, as shown in Columns 2 and 4, the models of Bharath and Shumway (2008) and Vassalou and Xing (2004) produce even larger estimation errors (on average 11%) suggesting that they do not lead to better estimates for total firm values. In contrast, the Eberhart (2005) parametrization produces a much lower mean estimation error (0.01) which is also much lower than the mean estimation error using book values. In fact, the obtained market-to-book ratios based on debt value estimates are very close to the observed bond market-to-book ratios.17 In summary, a considerable variation in market-to-book can be observed for tradable debt in our firm sample. The only parametrization which is capable of reproducing this variation and reduces also the estimation error compared to using book values of debt is based on the Eberhart (2005) model. To summarize: The Eberhart (2005) parametrization - is the most realistic parametrization ex ante (e.g. it uses a realistic time to maturity), - is the only parametrization which is able to replicate the cross sectional distribution of the true bond prices, and 16 Note that this procedure assumes that the outstanding bonds are representative for total firm debt, which is a simplifying but necessary assumption. 17 We also repeat the analysis using the absolute estimation error to assess the quality of the estimation procedures. In line with those results reported in Table 6, we find that the Eberhart (2005) application yields the lowest absolute estimation error (0.042). 18 - leads to the lowest estimation error. It is interesting to note that some parametrizations of the Merton (1974) model lead to higher estimation errors when compared to book values of debt. In the following, we therefore focus on a comparison between the excess value measure which uses book values of debt and an excess value measure which is based on market value estimates obtained by using the Eberhart (2005) assumptions. 5 Results Table 7 presents the median excess value measures for focused and diversified firms. Furthermore, the table contains the number of observations in each group as well as the p-value of a Wilcoxon rank-sum test (Mann-Whitney test). Null hypothesis is that the two samples are from populations with the same distribution. There are two main observations. First, replacing the book value of debt by market value of debt estimates leads to a reduction of the diversification discount. This is in line with our prediction and consistent with Mansi and Reeb (2002). When we use the Eberhart (2005) parameter values (which are the most plausible assumptions) to estimate market values of debt the discount is reduced from −15% to −5%, i.e. to one third of the original discount which is based on the traditional Berger and Ofek (1995) measure in which the book value of debt is used. Second, even when we use the Eberhart (2005) parameter values to estimate market values of debt, the resulting discount of −5% remains highly significant. This is contrary to Mansi and Reeb (2002). Our results suggest that the findings of Mansi and Reeb (2002), i.e. no significant differences, might be a result of the fact that the market values of debt they use are only available for about 13% of all firms in their sample, which dramatically reduces their number of observations. Table 8 shows coefficient estimates from pooled OLS (Regressions (1) and (2)) and fixed effects panel regressions (Regressions (3) and (4)) of excess value on a diversified firm indicator and control variables such as in Mansi and Reeb (2002). Excess value is the natural logarithm of the ratio of a firm’s actual value to its imputed value. A firm’s imputed value is the sum of the imputed values of its segments, with each segment’s imputed value 19 equal to the segment’s sales multiplied by its industry median ratio of capital to sales. Control variables are the natural logarithm of total assets, operating income divided by total assets, and capital expenditures divided by total assets. In Regressions (1) and (3), the excess value is based on the traditional Berger and Ofek (1995) measure which uses book values of debt. In Regressions (2) to (4), book value of debt is replaced by market value of debt estimates that are based on implementations of the Merton (1974) model suggested by Eberhart (2005). The results are similar to those presented in Mansi and Reeb (2002). Table 8 shows, for example, that operating income has a positive effect on the excess value. When the excess value is calculated using the traditional Berger and Ofek (1995) measure (Regressions (1) and (3)), the diversified dummy is significantly negative at least at the 5% level even in a regression with firm fixed effects. In Regression (4), the coefficient of the diversified dummy variable is only marginally significant at the 10 % level.18 In connection with the results of Table 7 we can conclude that replacing the book value of debt with market value of debt estimates leads to a reduction of the diversification discount. However, the discount does not completely vanish. The book value of debt bias, therefore, does not completely explain the diversification discount. 6 Discussion and Conclusion Prior literature documents that diversified firms sell at a discount relative to the sum of imputed values for their business segments. However, theoretical arguments suggest that corporate diversification has both a positive and a negative impact on shareholder value. Empirical research on the value impacts of corporate diversification suggests that firm value is decreasing in diversification. In this paper, we argue that there is actually much less evidence of a general loss of investor wealth associated with corporate diversification. Our major line of reasoning is that the excess value concept suggested by Berger and Ofek (1995) underestimates the firm value of diversified firms compared to focused 18 We also analyzed the robustness of our results. The main result is that the significance of the “diversified dummy” depends on the selection of other explanatory variables. In other words, firm fixed effects do not completely capture the diversification discount in Germany. 20 ones. And since most later studies which analyze potential causes of the diversification discount follow the Berger and Ofek (1995) procedure, they may also contain a systematic measurement bias and overestimate the discount. One of the obvious consequences of corporate diversification (apart from any potentially increasing inefficiencies in the internal capital budgeting process) is that it should lead to lower firm risk if business units are grouped together, which generate cash flows which are not perfectly positively correlated. In terms of the Merton (1974) model of debt valuation, the lower firm risk (lower volatility of the firm’s asset value) will increase the bondholder’s value at the expense of the shareholder’s value. The argument is that the equity of a firm can be viewed as a call option on the firm’s assets because shareholders are the residual claimants on the firm’s assets after any obligations have been met. Thus, there is reason to believe that book values of corporate debt are a more downward biased measure of market values for diversified companies. The traditional Berger and Ofek (1995) excess value measure compares the firm value of a company computed as the market value of equity plus the book value of debt to its imputed value if its segments operated as stand-alone firms. Book values of debt are used as a proxy for market values of debt because most corporate debt is not traded implying that market values are not observable. Under the assumption that book values are reasonable close to market values of debt and that any differences are not systematically related to the degree corporate diversification this approach is not problematic. However, this is unlikely to be the case as was described above. This argument was first put forward by Mansi and Reeb (2002), who compute the excess value for a subset of diversified firms using market values of debt. They find that there is no significant diversification discount if one relies on the market price of debt. We employ a different method and replace the book value of debt by market value of debt estimates that are based on often used implementations of the Merton (1974) model suggested by Bharath and Shumway (2008), Eberhart (2005), and Vassalou and Xing (2004). This procedure is possible for all firms in our sample. We are thus able to circumvent one weakness of the Mansi and Reeb (2002) who have to rely on data from about 13% of the usual sample of U.S. firms analyzed in other studies. 21 We find that different excess value measures are highly significantly positively correlated. In line with Mansi and Reeb (2002), the diversification discount is reduced but does not completely vanish. This hints to the point that the results presented in Mansi and Reeb (2002) might be driven by sample selection bias and lack of statistical power. We conclude that the book value of debt bias is only one explanation for the diversification discount. Future research should therefore analyze whether which economic factors are actually driving the diversification discount. Recently proposed factors are managerial optimism, managerial power, corporate governance, or efficiency of internal capital allocation which might be able to explain why some firms trade at a discount which are, in our regressions, captured by firm fixed effects so far. 22 References Amihud, Y., and B. Lev, 1981, “Risk Reduction as a Managerial Motive for Conglomerate Mergers,” Bell Journal of Economics, 12(2), 605–617. Ammann, M., D. Hoechle, and M. Schmid, 2008, “Is there Really no Conglomerate Discount?,” Working paper, Swiss Institute of Banking and Finance, University of St. Gallen,. Berger, P. G., and E. Ofek, 1995, “Diversification’s Effect on Firm Value,” Journal of Financial Economics, 37(1), 39–65. Bharath, S. T., and T. Shumway, 2008, “Forecasting Default with the Merton Distance to Default Model,” Review of Financial Studies, 21(3), 1339–1369. Borghesi, R., J. Houston, and A. Naranjo, 2007, “Value, Survival, and the Evolution of Firm Organizational Structure.,” Financial Management, 36(3), 5–31. Boston Consulting Group, 2006, “Managing for Value: How the World’s Top Diversified Com-panies Produce Superior Shareholder Returns,” BCG Report. Campa, J. M., and S. Kedia, 2002, “Explaining the Diversification Discount,” Journal of Finance, 57(4), 1731–1762. Campbell, J., J. Hilscher, and J. Szilagyi, 2008, “In Search of Distress Risk,” Journal of Finance, pp. –, forthcoming. Chevalier, J., 2004, “What Do We Know About Cross-subsidization? Evidence from Merging Firms,” Advances in Economic Analysis and Policy, 4(1), 1218–1218. Crosbie, P., and J. Bohn, 2003, “Modeling Default Risk,” Moodys KMV Company. Duffie, D., L. Saita, and K. Wang, 2007, “Multi-period corporate default prediction with stochastic covariates,” Journal of Financial Economics, 83(3), 635–665. Duffie, D., and K. J. Singleton, 2003, Credit Risk: Pricing, Measurement, and Management. Princeton University Press, Princeton, NJ. Eberhart, A. C., 2005, “A comparison of Merton’s option pricing model of corporate debt valuation to the use of book values,” Journal of Corporate Finance, 11(1-2), 401–426. Eom, Y. H., J. Helwege, and J.-Z. Huang, 2004, “Structural Models of Corporate Bond Pricing: An Empirical Analysis,” Review of Financial Studies, 17(2), 499–544. 23 Graham, J. R., M. L. Lemmon, and J. G. Wolf, 2002, “Does Corporate Diversification Destroy Value?,” Journal of Finance, 57(2), 695–720. Hellwig, M., C. Laux, and H. Müller, 2002, “Conglomeration: Good, Bad, or Unavoidable?,” Schmalenbach Business Review, special issue 1/2002, 55–78. Hill, C. W. L., and G. R. Jones, 2007, Strategic Management: An Integrated Approach. Houghton Mifflin, Boston, MA. Hyland, D., and J. D. Diltz, 2002, “Why Firms Diversify: An Empirical Examination,” Financial Management, 31(1), 51–81. Jensen, M. C., 1986, “Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers,” American Economic Review, 76(2), 323–329. , 1993, “The Modern Industrial Revolution, Exit, and the Failure of Internal Control Systems,” Journal of Finance, 48(3), 831–80. Lang, L. H. P., and R. M. Stulz, 1994, “Tobin’s q, Corporate Diversification, and Firm Performance,” Journal of Political Economy, 102(6), 1248–1280. Lee, K., M. W. Peng, and K. Lee, 2008, “From diversification premium to diversification discount during institutional transitions,” Journal of World Business, 43(1), 47–65. Lewellen, W. G., 1971, “A Pure Financial Rationale for the Conglomerate Merger,” Journal of Finance, 26(2), 521–37. Lins, K., and H. Servaes, 1999, “International Evidence on the Value of Corporate Diversification,” Journal of Finance, 54(6), 2215–2239. Lins, K. V., and H. Servaes, 2002, “Is Corporate Diversification Beneficial in Emerging Market?,” Financial Management, 31(2), 5–31. Lyandres, E., 2007, “Strategic Cost of Diversification,” Review of Financial Studies, 20(6), 1901–1940. Maksimovic, V., and G. Phillips, 2007, “Conglomerate Firms and Internal Capital Markets,” in Handbook of Corporate Finance, 1: Empirical Corporate Finance, ed. by B. E. Eckbo. forthcoming. Mansi, S. A., and D. M. Reeb, 2002, “Corporate Diversification: What Gets Discounted?,” Journal of Finance, 57(5), 2167–2183. Martin, J. D., and A. Sayrak, 2003, “Corporate diversification and shareholder value: a survey of recent literature,” Journal of Corporate Finance, 9(1), 37–57. 24 Merton, R. C., 1974, “On the Pricing of Corporate Debt: The Risk Structure of Interest Rates,” Journal of Finance, 29(2), 449–70. Meyer, M., P. Milgrom, and J. Roberts, 1992, “Organizational prospects, influence costs, and ownership changes,” Journal of Economics and Management Strategy, 1(1), 10–35. Rajan, R., H. Servaes, and L. Zingales, 2000, “The cost of diversity: The diversification discount and inefficient investment,” Journal of Finance, 55(1), 35–80. Santalo, J., and M. Becerra, 2008, “Competition from Specialized Firms and the Diversification-Performance Linkage,” Journal of Finance, 63(2), 851–883. Schaefer, S. M., and I. A. Strebulaev, 2008, “Structural models of credit risk are useful: Evidence from hedge ratios on corporate bonds,” Journal of Financial Economics, pp. –, forthcoming. Scharfstein, D. S., and J. Stein, 2000, “The Dark Side of Internal Capital Markets: Divisional Rent-Seeking and Inefficient Investment,” Journal of Finance, 55(6), 2537–2564. Servaes, H., 1996, “The Value of Diversification during the Conglomerate Merger Wave,” Journal of Finance, 51(4), 1201–25. Stein, J. C., 1997, “Internal Capital Markets and the Competition for Corporate Resources,” Journal of Finance, 52(1), 111–133. , 2003, “Agency, information and corporate investment,” in Handbook of the Economics of Finance, ed. by G. Constantinides, M. Harris, and R. M. Stulz. Elsevier, pp. 111–165. Stulz, R. M., 1990, “Managerial discretion and optimal financing policies,” Journal of Financial Economics, 26(1), 3–27. Vassalou, M., and Y. Xing, 2004, “Default Risk in Equity Returns,” Journal of Finance, 59(2), 831–868. Villalonga, B., 2004, “Does Diversification Cause the “Diversification Discount”?,” Financial Management, 33(2), 5–27. Weston, J. F., 1970, “Diversification and merger trends,” Business Economics, 5, 50–57. 25 Table 1: Number of Firms per Year This table presents the number of CDAX firms per year and the number of firms (ISINs) which have been in the CDAX between 2000 and 2006 with at least one piece of business segment information (assets, sales, or segment description for one segment not missing in given year; non-financial firms; SIC code between 6000 and 6999 deleted). The CDAX covers all German firms whose shares were admitted to the Prime Standard and General Standard segments of the German Stock Exchange. Before the introduction of these segments in the year 2003, the CDAX contained all firms whose shares were members of the segments Amtlicher Handel, Geregelter Markt, and Neuer Markt. ISIN means International Securities Identifying Number. Year Number of firms (ISINs) in CDAX (end of year) Number of firms (ISINs) which have been in the CDAX between 2000 and 2006 with at least one piece of business segment information (assets, sales, or segment description for one segment not missing in given year; non-financial firms; SIC code between 6000 and 6999 deleted) 2000 2001 2002 2003 2004 2005 2006 790 790 751 719 692 678 684 587 584 576 557 582 543 506 26 Table 2: Definition of Variables This table summarizes and defines the variables used in this paper. Variable Worldscope code Diversified firm (dummy) Description Firms with two or more operating segments which operate in different industries (based on two-digit segment SIC codes) Assets wc02999 Total assets Operating income wc01250 Capital expenditures wc04601 Total liabilities wc03351 Total debt wc03255 Current liabilities wc03101 Short term debt wc03051 Long term debt wc03251 Interest expense wc01251 Interest expense on debt Accounting standard (dummy) wc07536 Three dummy variables. Each dummy variable is set equal to 1 when a specific accounting standard is used and 0 otherwise. Firms are grouped into US-GAAP, local GAAP (HGB), and IFRS based on this Worldscope variable. 27 28 T r F E Risk-free rate Face value of debt Market capitalization σE Time to maturity Stock return (equity) volatility Input parameters market capitalization (stock price times shares outstanding) total debt 1-year Fibor 1 standard deviation of daily stock returns over the past 125 days Bharath and Shumway (2008) market capitalization (stock price times shares outstanding) current liabilities + 12 ·long term debt (KMV assumption) total debt·(1 + i)T with i =coupon rate (interest expense divided by total interest bearing debt) market capitalization (stock price times shares outstanding) 1-year Fibor 1 standard deviation of daily stock returns over the past 125 days Vassalou and Xing (2004) 1-year Fibor 0.6·short-term debt ratio +6.3·long-term debt ratio standard deviation of daily stock returns over the past 125 days Eberhart (2005) This table shows a summary of the input parameters used in this paper to estimate market values of debt. To estimate market values of debt, we solve equations (2) and (5), a system of nonlinear equations, for V and σV . Therefore, we need values for the remaining variables: Stock return (equity) volatility (σE ), time to maturity (T ), the risk-free rate (r), the face value of debt (F ), and the market capitalization (E). We use the set of input parameters suggested by three recent studies (Bharath and Shumway (2008), Eberhart (2005), and Vassalou and Xing (2004)). Table 3: Estimation of Market Values of Debt: Summary of Input Parameters Table 4: Bond Price Data: Descriptive Statistics This table shows the number of firms in our sample with tradable debt securities, separated by year and diversification degree and the relation of the bond amount outstanding to total respectively long-term firm debt. Year No. of firms focused No. of firms diversified Average of bond value in % of total debt [wc03255] Average of bond value in % of long term debt (%) [wc03251] 2000 2001 2002 2003 2004 2005 2006 0 0 1 3 1 1 2 17 23 42 58 49 59 65 28.0% 12.2% 21.7% 16.0% 13.1% 8.0% 8.3% 47.4% 17.3% 31.5% 24.0% 18.7% 10.8% 12.0% Total Mean 8 313 13.6% 20.0% 29 Table 5: Cross Sectional Distribution of Bond Prices and Market Value of Debt Estimates This table presents the cross sectional distribution of market-to-book values using observed bond prices and market value of debt estimates based on the parameterizations. Method Actual (bonds) Bharath and Shumway (2008) Eberhart (2005) Vassalou and Xing (2004) Mean Min p25 Median p75 Max N 1.08 0.97 1.09 0.97 0.95 0.95 0.93 0.95 1.03 0.96 0.98 0.96 1.07 0.97 1.06 0.97 1.10 0.98 1.12 0.98 1.46 0.98 1.38 0.98 321 318 318 318 30 Table 6: Estimation Error of the Market Value of Debt Estimates This table presents the estimation errors of the different approaches. See Section 4 for details on the calculation of the estimation error. Year Book values of debt Bharath and Shumway (2008) estimation Eberhart (2005) estimation Vassalou and Xing (2004) estimation 2000 2001 2002 2003 2004 2005 2006 0.03 0.05 0.08 0.09 0.11 0.09 0.04 0.08 0.08 0.11 0.11 0.14 0.12 0.08 -0.02 -0.05 -0.08 0.00 0.01 0.03 -0.01 0.08 0.08 0.11 0.11 0.14 0.12 0.08 Mean 0.08 0.11 -0.01 0.11 31 Table 7: Excess Value Measures for Focused and Diversified Firms This table presents the median excess value measures for focused and diversified firms. Furthermore, the table contains the number of observations in each group as well as the p-value of a Wilcoxon rank-sum test (Mann-Whitney test). Null hypothesis is that the two samples are from populations with the same distribution. *** indicates significance at 1%; ** indicates significance at 5%. EV measure based on Debt Berger and Ofek (1995) Berger and Ofek (1995) Book value Eberhart (2005) N Median Foc. N Median Div. p-value Wilcoxon 1411 1029 0 0 2375 2021 -0.147 -0.051 1.1e-08*** 0.019** 32 33 No Yes Yes 4070 0.031 Observations Firms (clusters) Adjusted R-squared R-squared within model R-squared overall model R-squared between model -0.139*** (0.000) 0.004 (0.509) -0.097 (0.336) -1.117*** (0.001) 0.059 (0.516) (1) 0.030 3118 No Yes Yes -0.082** (0.020) -0.001 (0.891) 0.182 (0.139) -1.327*** (0.001) 0.127 (0.278) Eberhart (2005) estimation (2) Pooled OLS Book value Firm fixed effects Year fixed effects Accounting standard fixed effects Constant Capital expenditures/total assets Operating income/total assets ln(total assets) Diversified firm (dummy) Debt measure Dependent variable: Excess value 0.020 0.014 0.003 4070 812 Yes Yes Yes -0.077** (0.014) -0.007 (0.561) 0.273*** (0.005) -0.769** (0.026) 0.234 (0.169) (3) Book value 0.032 0.017 0.007 3118 628 Yes Yes Yes -0.067* (0.087) 0.002 (0.885) 0.547*** (0.000) -1.001*** (0.008) 0.169 (0.422) Eberhart (2005) estimation (4) Panel regressions This table shows coefficient estimates from pooled OLS (Regressions (1) and (2)) and fixed effects panel regressions (Regressions (3) and (4)) of excess value on a diversified firm indicator and control variables such as in Mansi and Reeb (2002). Excess value is the natural logarithm of the ratio of a firm’s actual value to its imputed value. A firm’s imputed value is the sum of the imputed values of its segments, with each segment’s imputed value equal to the segment’s sales multiplied by its industry median ratio of capital to sales. Control variables are the natural logarithm of total assets, operating income divided by total assets, and capital expenditures divided by total assets. In Regressions (1) and (3), the excess value is based on the traditional Berger and Ofek (1995) measure which uses book values of debt. In Regressions (2) to (4), book value of debt is replaced by market value of debt estimates that are based on implementations of the Merton (1974) model suggested by Eberhart (2005). Time period is 2000 to 2006. Year dummies and accounting standard dummies are included. See Table 2 for details on the definition of variables. Variables are winsorized at the 1% level. Robust p-values are in parentheses. * indicates significance at the 10% level, ** indicates significance at the 5% level; *** indicates significance at the 1% level. Table 8: Determinants of the Excess Value